There’s a certain kind of developer who tabs through their monthly credit card statement and quietly calculates how much of it is going to AI subscriptions. Claude Max. Cursor Pro. GitHub Copilot. Maybe a separate API key they justified as “experimental.” It adds up.

And then they hear someone mention the GLM Coding Plan at $18 a month and immediately assume something is broken. Either the benchmarks are fabricated, the rate limits are punishing, or the thing only works in Mandarin. None of that is true. But the skepticism is reasonable, because $18 for frontier-class coding AI does sound too good.

So let’s actually pull it apart. What you get, what the limits look like in practice, how ZCode fits into the picture, and where the real catches are.

What the GLM Coding Plan Is (and Isn’t)

The GLM Coding Plan is Z.ai’s flat-fee subscription for developers who want to use GLM models inside their preferred coding tools. Instead of burning through API tokens at per-million-token rates, you pay a fixed monthly amount and get a fixed quota of prompts.

It works by giving you access to Z.ai’s Anthropic-compatible API endpoint. You point your coding tool at Z.ai’s servers, enter your plan credentials, and your chosen tool uses GLM-5.2 (or another model in the plan’s lineup) as the underlying model instead of Claude, GPT-5.5, or whatever you were using before.

The plan isn’t a standalone product with its own interface. You’re not switching editors. You’re switching the model provider underneath your existing setup. Claude Code, Cline, Roo Code, Kilo Code, OpenCode, Goose, OpenClaw, and 20+ other tools all work. If your current tool supports custom model providers, you’re almost certainly compatible.

This is worth repeating because it confuses a lot of people early on: you still use whatever coding tool you already use. The plan just changes what model that tool calls.

The Tiers, Broken Down Honestly

As of mid-2026, there are three main tiers. Prices below reflect standard monthly rates; annual billing knocks 30% off, and a 10% discount applies to monthly billing (yes, even monthly billing is discounted).

Tier	Standard Monthly	Annual Rate (per month)	Prompts per 5 hours	Weekly Cap	MCP Calls
Lite	$18/mo	~$12.60/mo	~80 prompts	~400/week	100/month
Pro	$72/mo	~$50.40/mo	~400 prompts	~2,000/week	1,000/month
Max	$160/mo	~$112/mo	~1,600 prompts	~8,000/week	4,000/month

All three tiers include the full model lineup: GLM-5.2 (the flagship), GLM-5-Turbo, GLM-4.7, and GLM-4.5-Air. They also include Vision Understanding, Web Search MCP, Web Reader MCP, and Zread MCP. You don’t get a stripped-down version of the model at Lite and a better one at Max. You get the same models; you just get more quota.

The Pro tier is roughly 5x Lite on usage. Max is roughly 20x Lite. If you code heavily all day, Lite will run out fast. Pro is what most solo developers actively doing agentic workflows will want. Max is for teams or individuals who are running long autonomous agents across large codebases for hours at a time.

One practical note before we go further: start on Lite with monthly billing. Use it for a real month. Then lock in annual once you know what tier your actual usage demands. The discount is worth it, but so is not committing $1,344 to a plan before you know you’ll use it.

Pricing reflects publicly available rates as of July 2026 and varies by billing cycle and active promotions. Always verify at z.ai before subscribing.

The Quota System (Where It Gets Complicated)

The part that trips people up is the multiplier system, and it’s worth understanding before you subscribe.

One “prompt” equals one user turn in your coding tool. When your coding agent calls the model, that counts as a prompt. The actual number of times GLM-5.2 gets called internally per prompt ranges from roughly 15 to 20, but for quota purposes, it’s all one user-initiated call.

Here’s where the multipliers kick in:

During peak hours (14:00 to 18:00 UTC+8, which is 06:00 to 10:00 UTC), GLM-5.2 and GLM-5-Turbo consume quota at 3x the standard rate.

During off-peak hours, it’s normally 2x. However, a promotion running through September 2026 drops off-peak to 1x, meaning your quota stretches much further if you’re working outside that peak window.

GLM-4.7 and GLM-4.5-Air are cheaper to run and don’t trigger the same multipliers. The practical strategy most experienced Coding Plan users land on: use GLM-4.7 for routine tasks (autocomplete, small fixes, docstring generation), save GLM-5.2 for the complex stuff (large refactors, architecture decisions, multi-file changes). This keeps your quota sustainable.

If you hit your weekly cap, the plan stops. There’s no auto-upgrade, no overage billing, and no grace period. It just stops until the next quota reset. So knowing your usage pattern before committing to a tier matters.

Is It Cheaper Than Paying Per Token?

For most developers doing continuous agentic coding work all day, yes. By a lot.

Here’s the rough math. On the standard Z.ai API, GLM-5.2 runs $1.40 per million input tokens and $4.40 per million output. A single complex coding prompt with a large codebase context can easily consume 20,000 to 50,000 tokens. Multiply that across a full workday and you’re spending $50 to $100+ per day on API costs alone.

The Coding Plan’s monthly equivalent on API billing: roughly $150 per month at the value the plan quotes (15 to 30x the subscription fee in token equivalent), based on Z.ai’s own documentation estimates.

API billing wins when your usage is bursty or programmatic (scripts that make occasional calls, automated pipelines that run a few times a day). Coding Plan wins when you’re sitting in an agentic IDE for hours, asking the model to do real multi-step work.

The comparison against other subscription tools tells a similar story:

Tool	Price	Model	Monthly Usage
GLM Coding Plan Lite	$18/mo	GLM-5.2	~400 prompts/week
GLM Coding Plan Pro	$72/mo	GLM-5.2	~2,000 prompts/week
Claude Code (individual)	~$100/mo	Claude Sonnet/Opus	Varies by plan
Cursor Pro	$20/mo	Various	Fixed credits
Cursor Ultra	$200/mo	Various	Higher credits
GitHub Copilot	$19/mo	Various	Monthly requests

At $18 for a Pro-equivalent experience (depending on which Claude Code tier you’re comparing against), the GLM Coding Plan Pro at $50.40/month on annual billing sits well below comparable quotas from Western alternatives for developers who find GLM-5.2’s performance acceptable for their workload.

What ZCode Is and Why It’s Different

ZCode is Z.ai’s answer to Cursor. Or to Claude Code. Depending on how you frame it.

Officially, Z.ai describes it as an Agentic Development Environment (ADE) rather than an IDE. The distinction is deliberate and actually meaningful. Where Cursor or a VS Code fork puts the editor at the center and adds AI incrementally, ZCode puts the agent loop at the center and arranges everything else around it. The file manager, the terminal, the Git panel, and a live browser preview are all there, but they’re secondary to the agent conversation.

ZCode launched publicly the week of July 1, 2026, available as a free download for macOS, Windows, and Linux (Linux still in beta as of launch). The app itself is free. You pay for the AI models you connect.

The core abstraction is /goal. Type something like /goal refactor authentication module to use JWT and add email-based MFA and the agent plans the work, writes the code, runs tests, monitors terminal output, and keeps iterating until the goal is verified complete or it hits a blocker it can’t resolve. A status panel tracks elapsed time, token consumption, and iteration count throughout. You can pause, redirect, or clear a goal mid-task without losing session state.

This is the “agentic” part that makes it different from a chat sidebar. You’re not typing in a prompt and reviewing the output. You’re setting an objective and the agent executes it autonomously. That shift in workflow is either exciting or terrifying depending on how much you like being in the loop.

ZCode Key Features

Long-task execution. The agent maintains context across files, terminal output, browser state, Git state, and execution modes throughout a single task. This is where GLM-5.2’s 1-million-token context window becomes practically useful rather than just a marketing number.

Five execution modes. ZCode ships with multiple operation modes (cycled with Shift+Tab) that control how autonomously the agent operates. Less autonomous modes require confirmation before sensitive commands; fully autonomous modes can run without interruption. This dial matters, especially after a few high-profile incidents of agents chaining actions without human review.

Custom subagents. Defined as plain Markdown files at ~/.zcode/agents/, subagents are scoped mini-agents for specific tasks. You define a name, model, available tools, and a description of what it does. Call it with /agent test-runner and it runs your test suite, returns a summary of failures with line numbers, and that’s it. Highly targeted, less likely to wander.

Remote and bot control. You can start a long-running task on your desktop and monitor it from your phone. There’s also integration with WeChat, Feishu, and Telegram bots, which reveals plainly who the primary intended audience is for now, though the remote capability is useful regardless of your preferred messaging platform.

SSH and Docker support. Point ZCode at a remote machine or container and the agent executes there. Useful for testing against production-like environments without needing to clone large repos locally.

GLM Coding Plan quota promo. Through July 31, 2026, Coding Plan subscribers using ZCode get a quota discount: peak hours drop from 3x to 2x, effectively giving you about 1.5x the usable quota inside ZCode compared to using the plan in another tool. After the promo expires, standard multipliers apply.

First-time ZCode users get 5 days of free usage with no setup required: 3 million GLM-5.2 tokens plus 2 million GLM-5-Turbo tokens per day. Enough to run the agent against a real repository and form a genuine judgment before subscribing to anything.

ZCode vs Claude Code: An Honest Comparison

People keep framing this as one replacing the other. It’s more complicated.

Factor	ZCode	Claude Code
Interface	Agent-first ADE (desktop app)	Terminal CLI / VS Code extension
Primary model	GLM-5.2 (via Coding Plan)	Claude Sonnet/Opus (Anthropic)
Open weights	Yes (MIT)	No
Context window	1M tokens	200K tokens
Hardest SWE tasks	Slightly behind Opus 4.8	Better on longest-horizon tasks
Data routing	China-based servers (cloud)	US-based Anthropic servers
Self-hosting option	Yes (full open weights)	No
BYOK support	Yes	Yes (external providers)
Price	GLM Plan from $18/mo	From ~$20/mo (varies by tier)
Best for	Budget-conscious teams, open-weight requirement	Hardest agentic work, Anthropic ecosystem

The honest read is that Claude Opus 4.8 is still the ceiling for the most complex, sustained, long-horizon engineering tasks. If you’re working on something where those final few percentage points of SWE-bench performance translate to real task completion vs. partial completion, Opus 4.8 is still the pick.

For the wide middle of actual development work, GLM-5.2 via the Coding Plan is a competitive option at a materially lower price. The 7-point gap on SWE-bench Pro is real. Whether it matters to your specific workflow is something only real usage will tell you.

The Real Catches

There are three things worth being clear-eyed about before subscribing.

The data routing issue. Covered in the Z.ai overview post, but worth repeating: cloud API calls go through Z.ai’s China-based infrastructure. This is a non-issue for most personal projects and startups. For enterprises in regulated industries, it requires either a legal review or a shift to self-hosting.

The quota multiplier after the promo expires. Right now, off-peak GLM-5.2 usage is at 1x quota consumption through September 2026. After that, it goes back to 2x off-peak and 3x peak. If you’ve built your workflow around current consumption rates, you’ll want to reassess your tier once the promo lifts.

ZCode is agent-first, not editor-first. If your muscle memory is built around tab-complete, sidebar suggestions, and inline diffs, ZCode will feel wrong for a while. It’s not designed to be a better Cursor. It’s designed to replace the loop where you’re the one breaking down the task and the agent handles the execution. If that’s not how you work, the learning curve is real.

Frequently Asked Questions About the GLM Coding Plan and ZCode

Final Thoughts

The GLM Coding Plan is a real product with real performance at a price that makes sense. It’s not a charity offer. Z.ai is a well-capitalized lab with a clear commercial strategy, and the pricing is aggressive because they want market share during a critical window.

For developers who are spending $100+ a month on AI coding subscriptions and getting frustrated by rate limits, the Coding Plan is worth a serious trial. The 5-day ZCode free trial costs nothing and will tell you more than any benchmark comparison ever will.

The catch is that “worth it” is contextual. If you’re working on extremely long-horizon tasks in a large enterprise codebase where the Claude Opus 4.8 gap in SWE-Marathon actually matters, this is a budget option, not a replacement. If you’re an indie developer building a SaaS product and hitting your Claude credits mid-month, the math works strongly in the Coding Plan’s favor.

Try it against something you’d actually build. That’s the only benchmark that matters for your use case.

Pricing estimates in this article reflect publicly available rates as of July 2026 and will vary by billing cycle, region, and active promotions. Always verify current pricing at z.ai/subscribe before making a decision.

---

Curated by Lorphic
Digital intelligence. Clarity. Truth.

Meta Title: GLM Coding Plan & ZCode Explained: Pricing, Tiers & Worth It? (2026) Meta Description: GLM Coding Plan offers GLM-5.2 in Claude Code, Cline, and 20+ tools from $18/mo. Full breakdown of tiers, ZCode IDE, quota limits, and when it beats Claude Code. Slug: glm-coding-plan-zcode Focus Keyword: GLM coding plan Secondary Keywords: ZCode z.ai, z.ai coding plan pricing, GLM-5.2 coding